Symptom-based phenotype discovery in motor neuron disease using natural language processing of electronic health records

A research team developed a natural language processing system to extract symptoms from clinical notes of 2,361 patients with confirmed motor neuron disease diagnosis. From initial clinical notes, they identified four patient groups with specific symptoms: motor-bulbar, motor-tremor, sensory-pain, and motor-respiratory. When analyzing all available clinical notes, these reorganized into three clusters: autonomic-respiratory, nocturnal-respiratory, and classic motor. Survival differences were statistically significant among all groups. The results show that autonomic-respiratory symptoms are associated with worse prognosis, while the newly identified sensory-pain subtype has better prognosis. These data-driven phenotypes may improve disease progression prediction and inform targeted supportive care.